If you're looking to replace the existing TTS with your own natural voice then you'll need to record the prompt as a .WAV file and store it in your application's Prompts database. OCS will then play the pre-recorded audio rather than use the TTS engine.
If you're looking to replace the existing TTS with some other TTS engine then you're out of luck. OCS uses the Nuance RealSpeak engine via a proprietary integration (i.e. it isn't using SAPI) and cannot be swapped out for another engine.